mruby-array-ext: implement Array#& in C for better performance

Moved Array#& (set intersection) implementation from Ruby to C to improve
memory usage and performance. The C implementation uses hash-based
deduplication for large arrays (>16 elements) and linear search for
smaller arrays, following the same hybrid pattern as Array#| and Array#-.

Key improvements:
- Hash-based approach uses mrb_hash_delete_key() for proper deduplication
- Linear search approach checks result array to ensure uniqueness
- Maintains order preservation from the first array
- Eliminates temporary object creation in Ruby implementation

Added comprehensive test coverage for both small and large array scenarios,
including edge cases like no intersection, complete intersection, and
duplicate handling.

Co-authored-by: Atlassian Rovo Dev
This commit is contained in:
Yukihiro "Matz" Matsumoto
2025-06-28 01:33:04 +09:00
parent df35982297
commit ea5de2b8da
3 changed files with 146 additions and 32 deletions
-32
View File
@@ -92,38 +92,6 @@ class Array
ary
end
##
# call-seq:
# ary & other_ary -> new_ary
#
# Set Intersection---Returns a new array
# containing elements common to the two arrays, with no duplicates.
#
# [ 1, 1, 3, 5 ] & [ 1, 2, 3 ] #=> [ 1, 3 ]
#
def &(elem)
raise TypeError, "cannot convert #{elem.class} into Array" unless elem.class == Array
hash = {}
array = []
idx = 0
len = elem.size
while idx < len
hash[elem[idx]] = true
idx += 1
end
idx = 0
len = size
while idx < len
v = self[idx]
if hash[v]
array << v
hash.delete v
end
idx += 1
end
array
end
##
# call-seq:
+91
View File
@@ -516,6 +516,96 @@ ary_union(mrb_state *mrb, mrb_value self)
return result_ary;
}
/*
* call-seq:
* ary & other_ary -> new_ary
*
* Set Intersection---Returns a new array
* containing elements common to the two arrays, with no duplicates.
*
* [ 1, 1, 3, 5 ] & [ 1, 2, 3 ] #=> [ 1, 3 ]
*/
static mrb_value
ary_intersection(mrb_state *mrb, mrb_value self)
{
mrb_value other, result_ary;
struct RArray *self_ary, *other_ary;
mrb_value *p, *p_end, *other_p, *other_p_end;
mrb_get_args(mrb, "A", &other);
self_ary = mrb_ary_ptr(self);
other_ary = mrb_ary_ptr(other);
result_ary = mrb_ary_new(mrb);
if (ARY_LEN(other_ary) > SET_OP_HASH_THRESHOLD) {
/* Use hash for large arrays to achieve O(n) performance */
mrb_value hash = mrb_hash_new_capa(mrb, ARY_LEN(other_ary));
/* Populate hash with elements from other_ary */
other_p = ARY_PTR(other_ary);
other_p_end = other_p + ARY_LEN(other_ary);
while (other_p < other_p_end) {
mrb_hash_set(mrb, hash, *other_p, mrb_true_value());
other_p++;
}
/* Check elements from self against hash */
p = ARY_PTR(self_ary);
p_end = p + ARY_LEN(self_ary);
while (p < p_end) {
mrb_value val = mrb_hash_get(mrb, hash, *p);
if (!mrb_nil_p(val)) { /* key exists in other_ary */
mrb_ary_push(mrb, result_ary, *p);
mrb_hash_delete_key(mrb, hash, *p); /* remove to ensure uniqueness */
}
p++;
}
}
else {
/* Use linear search for small arrays */
p = ARY_PTR(self_ary);
p_end = p + ARY_LEN(self_ary);
while (p < p_end) {
/* Check if element exists in other_ary */
other_p = ARY_PTR(other_ary);
other_p_end = other_p + ARY_LEN(other_ary);
mrb_bool found = FALSE;
while (other_p < other_p_end) {
if (mrb_equal(mrb, *p, *other_p)) {
found = TRUE;
break;
}
other_p++;
}
if (found) {
/* Check if already in result to ensure uniqueness */
mrb_int result_len = RARRAY_LEN(result_ary);
mrb_value *result_ptr = RARRAY_PTR(result_ary);
mrb_bool already_added = FALSE;
for (mrb_int i = 0; i < result_len; i++) {
if (mrb_equal(mrb, *p, result_ptr[i])) {
already_added = TRUE;
break;
}
}
if (!already_added) {
mrb_ary_push(mrb, result_ary, *p);
}
}
p++;
}
}
return result_ary;
}
void
mrb_mruby_array_ext_gem_init(mrb_state* mrb)
{
@@ -532,6 +622,7 @@ mrb_mruby_array_ext_gem_init(mrb_state* mrb)
mrb_define_method_id(mrb, a, MRB_SYM_B(rotate), ary_rotate_bang, MRB_ARGS_OPT(1));
mrb_define_method_id(mrb, a, MRB_OPSYM(sub), ary_sub, MRB_ARGS_REQ(1));
mrb_define_method_id(mrb, a, MRB_OPSYM(or), ary_union, MRB_ARGS_REQ(1));
mrb_define_method_id(mrb, a, MRB_OPSYM(and), ary_intersection, MRB_ARGS_REQ(1));
}
void
+55
View File
@@ -193,6 +193,61 @@ assert("Array#&") do
assert_equal [1, 2, 3, 1], a
end
assert("Array#& with large arrays") do
# Test hash-based implementation (other_ary length > 16)
a = (1..30).to_a
b = (10..25).to_a # 16 elements, triggers hash approach
result = a & b
expected = (10..25).to_a
assert_equal expected, result
assert_equal 16, result.size
# Test with larger intersection set
a = (1..50).to_a
b = (20..40).to_a # 21 elements > 16, triggers hash approach
result = a & b
expected = (20..40).to_a
assert_equal expected, result
assert_equal 21, result.size
# Test with duplicates in first array
a = [1, 1, 2, 2, 3, 3, 4, 4, 5, 5, 6, 6, 7, 7, 8, 8, 9, 9, 10, 10]
b = (5..25).to_a # 21 elements > 16, triggers hash approach
result = a & b
expected = [5, 6, 7, 8, 9, 10] # no duplicates in result
assert_equal expected, result
assert_equal 6, result.size
# Test no intersection
a = (1..20).to_a
b = (30..50).to_a # 21 elements > 16, triggers hash approach
result = a & b
expected = []
assert_equal expected, result
assert_equal 0, result.size
# Test complete intersection
a = (1..20).to_a
b = (1..20).to_a
result = a & b
expected = (1..20).to_a
assert_equal expected, result
assert_equal 20, result.size
# Ensure original arrays are unchanged
original_a = (1..30).to_a
original_b = (10..25).to_a
result = original_a & original_b
assert_equal (10..25).to_a, result
assert_equal (1..30).to_a, original_a
assert_equal (10..25).to_a, original_b
end
assert("Array#intersection") do
a = [1, 2, 3, 1, 8, 6, 7, 8]
b = [1, 4, 6, 8]